How to Quiet a Noisy Dragon

My last post about my quick test with the new Dragon sensor had a bit of a surprise with the Dragon footage looking noisy, especially compared to the previous non-Dragon MX chip.  RED suggested that a fix was on the way, and soon.  They hinted at around a week.  That was June 21st.  My rule of thumb for RED target dates for delivering a product is: take the stated time period, double it, and add two months.  So, in this case that would mean somewhere around the first week of September. Well, RED beat the odds, while still completely missing their target of a week, and August 6, 2014, a mere 7 weeks, they released the fix.   The fix is a different way to debayer the RAW footage, selectable in their new beta release of REDCINE. The feature is called “DEB” or “Dragon Enhanced Blacks” although it could easily be called the “Anti Red Speckle Filter” as it gets rid of the red noise. it is a checkbox you select right below the Gamma settings in REDCINE.

DEB checkbox just below Gamma Settings in REDCINE

DEB checkbox just below Gamma Settings in REDCINE.  Here it is not selected, the current default.

The good news is it is retroactive and can be applied to footage you have already shot.  This is great for my purposes as I don’t have t re-do the test.  Here are some quick screen grabs.

same shot from prior test, with DEB allied and not.

same shot from prior test, with DEB applied and not.

DEB dragon Vs Epic-X

RED Epic MX on left, Lower right is Dragon with DEB applied.

So it definitely improves things. Further testing is warranted though. I hope to test it against a different camera, like Arri Alexa or Amira in the future.  RED is talking about making user swappable OPLFs but no time frame on that yet.  Once they do put a date on it, don’t forget to double it and add two months!

Epic MX vs Epic Dragon.

I just recently got my RED Epic back after RED installed the new “Dragon” chip.  I borrowed an Epic that still had the MX chip and shot side by side tests to see how much better the Dragon chip actually was.  I found a few surprises.

First off, I updated both cameras to the current release builds 5.1.51 and black shaded both of them after cameras had reached operating temp at adaptive at 65c.

IMG_4748the weather was a bit unsettled (as it always seems to be when I seem to have time to do these tests) so I decided to put on a 35mm Red Pro prime on the Epic MX and a 50mm RPP on the Dragon and move the cameras a bit to mimic same frame size of midground so I could roll both at the same time so exposure would be identical.  I set them for the same stop, which if I recall was something like f 8 1/3.  according to the false color overlay the Dragon had more info in the highlights before clipping.  In fact a small patch of white sky which clipped on the MX was apparently not clipping in the Dragon. I shot myself in my garage workspace that has diffused top light and a nice window for testing overexposure.  If I shoot in the afternoon there is a piece of an apartment building across the street that gets hit with sun, providing an excellent detail/ overexposure test element.  People have complained in the past about RED’s rendition of skin tones.  I think this is wildly overblown.  I have never had an issue with it, at least under daylight color temps, which is how I shoot the majority of my stuff.  The Dragon is supposed to be much better.  I did not notice much of a difference, nor did I see any problems with the old one.  One note is that with the new chip comes some new settings in REDCine, the app that lets you “develop” the Raw footage from the camera.  before Dragon there was “Redcolor 3” for color rendition, and “REDgamma 3” for gamma.  With Dragon there now is “Dragon Color” and “REDgamma 4.”  while I think you should use Dragoncolor for rendition off a dragon chip, don’t be fooled into thinking REDgamma 4 is automatically better.  It tends to be a bit crunchier than REDgamma 3 which under normal circumstances helps make a punchier image, but when testing over under exposure range, may fool you into thinking the Dragon chip has even less range than the Regular Epic.  Of course, professional graders probably would just use REDlogfilm (which is flat and holds onto the most range) and make their own curve based on the scene. but in this case I wanted to do a relatively unbiased side by side test.

screenshot from REDcine.

screenshot from REDcine.

Below are some interesting screen grabs:

COLOR RENDITION:

Side by Side Dragoncolor Vs REDcolor3

Side by Side Dragoncolor Vs REDcolor3

Dragon on the right, Epic-MX on left. Dragon set to Dragoncolor Epic MX set to REDcolor3  More detail in Dragon, and slightly less red tint to the skin.

HIGHLIGHTS:

Now here is a detail of the highlights: Dragon on the right, Epic on the left.  Both set to REDgamma3. slightly more highlight detail on the Dragon chip as expected.

RG3 Dragon vs MX
Don’t be fooled into using REDgamma 4 just because it is better.DRAGON RG4 VS EPIC RG3 Here is the same footage

but Dragon chip is rendered at REDgamma4.  It is very hard to tell which chip has the advantage in this scenario.brick looks about the same, maybe slight advantage Dragon, but leaves look hotter on the Dragon.

Now here is the same thing in REDlogfilm. You will see the Epic MX chip clips magenta showing the white is clipping.  Not so in the Dragon. as you might expect.

Epic-MX vs Dragon in REDlogfilm

Epic-MX vs Dragon in REDlogfilm

BIG SURPRISE: NOISE LEVEL:

This is Epic MX on left and Dragon on the right

This is Epic MX on left and Dragon on the right. both at 5600K, 800 ISO, 8:1 compression.  Dragon at 6KHD and Epic-MX at 5KHD.

 

This was the big surprise. The Dragon Epic was sharper, as was to be expected as it’s max resolution was better than the MX (6KHD in this case vs 5KHD) but there was more noise and pattern in the Dragon footage.  This was alarming as a year ago the Dragon chip had been advertized as 2000 ISO was as quiet as 800 ISO in the MX chip. What the hell? I paid over 8 grand to get MORE noise?

I went over to REDUSER.net and fund someone else had done essentially the same test as I had done and got the same results. if you are into flogging yourself, here is the link:

http://www.reduser.net/forum/showthread.php?117701-DRAGON-vs-EPIC-MX-Noise-NOT-GOOD

CONCLUSION:

Anyway the long and short of it was this: It is the new OPLF, along with black shading that is to blame.  When Dragons first started shipping to anybody (maybe December 2013?) they had what I will call version 1 of the OPLF (Optical Low Pass Filter) which is essentially a piece of custom class/filter in front of the sensor proper to improve performance, reject IR and limit moire.  All cameras have some kind of OPLF.  after logging some hours with it in the hands of users the following characteristics were found.  It did in fact seem pretty clean at 2000 ISO.  But there were weird magenta flares, which were very visible in low light situations.  A new OPLF was designed that got rid of the magenta flare, improved highlight performance by at least a stop (I have no idea how they did that, possibly in conjunction with new black shading algorithm?) and even better IR performance.  Downside? now 2000ISO is noisy.  And the whole thing blew up on REDUSER once everyone was getting new Dragons in numbers with the new OPLF.

The good news is that RED will, if you wish, put the old OPLF back in your camera, if that is what you wish.  And they claim to have a firmware build in the works (Don’t they always)  that will tweak black shading implementation that will address this problem, and that it will come out soon, possibly in a week.  But knowing RED that could mean a month.  They claim that the new OPLF brings so much to the table that they think the future is in the new OPLF with software tweaks. There even has been talk of user replaceable OPLFs in the future. I can confirm that the IR contamination on this new arrangement is quite good.  I had a shoot with an n1.2 and pola (6 stops of ND essentially) and a black dress Marine uniform in full sun and the uniform stayed black.  unfortunately I had shot it at 2000 ISO because I had believed that 2000 was the new 800, and that shooting at a higher ISO protected your highlights more. This had been true of the RED One, but is no longer true if RED Dragon.

Also, I rendered both the Dragon and the MX footage to a 1080 ProRes file.  the noise difference was indiscernible, which was a pleasant surprise.  I did not use any special noise reducing options on either.  So for the moment I am content to wait for the new firmware and black shading.  Even if it makes all the testing above obsolete.  But as always, test test test.  Sometimes you find a surprise.

 

ProRes Camera Test: BMPCC, F3, F5, and RED Epic MX

Recently I decided to test some of the cameras I often use and or have access to.  This included the Blackmagic Pocket Cinema Camera, a Sony F3, a Sony F5, A RED Epic MX.  Obviously these cameras record natively in different codecs and resolutions, but I decided to even the playing field somewhat by using an external recorder (A Convergent Design Odyssey 7Q) so that everything is recorded in ProRes HQ, except for the Pocket Cinema Camera, which both records ProRes HQ natively, and also only outputs via HDMI (Highly Dodgy Media Interface) which I hate and prefer not to use. In all the tests I used the same lens, a RED 18-50mm short zoom, which was PL mount so it fit on all the cameras, which all had PL mounts on.  Only for the Pocket Cinema camera did that significantly change the filed of view, as it is more of a super 16mm sensor size compared to 35mm sensor size of the others, which I compensated by zooming out to approximate the same field of view.  The RED Epic of could natively can record up to 5K, but in this case I was just recording the 1080 output from a 4KHD recording, as 4KHD most closely matches 35mm format size.  Also I did not record LOG on the F3 or F5, or output REDLog on the Epic.  I did do both “Film” Rec and “Video” modes in the Pocket Cinema camera, but I did not run them through Resolve and color correct them as I might if I were actually using the camera.  I wanted to look at the footage without doing any alterations other than what I might do in camera.  For the F3 and F5 which have scene files that affect the look of the camera, I arbitrarily picked Abel Cine’s JR45Cine scene file for both, anticipating that those would be the scene files I would use were I to shoot with these cameras.  For the Epic, I usually shoot with saturation set to 1.2 rather than the default 1.0, and that is what I did here. White balance was preset 5600 for all cameras. it was a somewhat unsettled day weather-wise, so take any ability to hold detail outside the window with a grain of salt as these shots were not all done simultaneously, as I was using one lens for all the cameras, and I had only two tripods in play.

Click on any below for full frame.

No surprises here.  Epic and F5 look the best.  the F5 skin tones look a bit too red, but the whites look a little truer I think. the Black Magic Pocket Cinema Camera looks like it needs to be graded some whether you shoot video or film mode, and therefore makes me think if I am going to have to grade it anyway to make it look its best, I would shoot in “Film” mode in the future and take it through Resolve.

Below are enlargements of the focus target. again, no surprises.  Sony F5 and RED Epic are the best for resolving and lack of moire.  I was surprised slightly at the F3 moire, and not surprised but slightly disappointed on the Black Magic Pocket Cinema Camera’s sharpness performance. Again, this is off a screen grab from the ProRes HQ 1080 recording.

If I were to pursue this test further, I would explore scene files in the F3 and F5 to see if I could get truer skin tones out of it while maintaining their ranges.  The RED I might play with white balance a little bit, maybe a little lower to get rid of some of the tones that seem a bit too warm.  My next test I had planned though was to test my RED Epic after it had it’s sensor upgraded to the “Dragon” sensor.

Jeffrey Brown’s eyeball

cu wideThe current almost beaten to death fad in cameras these days is trying to get a shallow depth of field, i.e. shallow focus look.  Of course people are crazy about it because it looks great for portrait stuff or when run by a professional crew that can maintain the critical focus throughout the shot.  But when budgets get tight, and someone whips out a Canon 5D instead of a true video or film camera, and isn’t properly staffed, you end up with constant hunting for focus and lots of blurry shots and lots of justification that “it’s a look.”

In controlled environments it is much less of an issue, or so I thought.  I recently shot some last minute promos for PBS News hour.  They were to be the various correspondents against black limbo delivering one line while standing. Budget was of course tight, but these were really just head shots, nothing too complicated.  The director was interested in using primes, as they tend to be sharper and tend to be able to give you shallower depth of field due to faster f stops.  Fine by me.  Of course, since it is black limbo, any shallow depth of field look you are after will have to play out exclusively on their face as there is literally nothing else to see in the frame. Mainly I was interested in the sharpness that primes could provide.  They were interested in both 4K acquisition, so we shot on my RED Epic camera using two RED Pro Primes, a 50mm and a 100mm. Again, nothing too complicated.  We set up the lighting, and had the talent come in and do their bit to camera.

At 100mm at just under f2.8 focus was razor-thin.  Even with the talent standing still, keeping it in focus was very hard.  I did not have a First Assistant, partially because of budget and because I had not pushed for one, as everything was static, and I had not seen a need for one.  I did have someone to deal with the data downloading, but no one skilled in pulling focus except me. I ended up staring at a 17″ monitor from as close as I could get, with my visual world shrinking down to the key side eyeball and trying to keep it sharp.  One consistent thing I noticed is that even standing still people will lean forward ever so slightly upon delivering a line, as a sort of body language emphasis.  It is something you don’t normally notice, unless you have less than a 1/4 inch of useable focus. About three people in we had Jeffrey Brown as talent.

A002_C007_0424ZZ.0002435 In addition to being a good journalist he has very distinctively slate blue eyes.  Everyone always looks to the eyes for focus, but with his eyes it seemed especially critical as they really “snapped” when in focus, which of course meant that they went soft, even for a moment everybody would notice.  And he was a “rocker,” leaning just a bit more than some of the other talent, making focus an all-consuming job.  At one point the director had to point out that the framing of the shot was going to hell, as I had ceased to be monitoring it, devoting all my time to keeping that damned eyeball in focus.  I apologized for the poor framing saying, “I was distracted by the eyeball” which got a few laughs, especially since Jeffrey Brown thought it was because it was because of his blue eyes, which he is somewhat sensitive to as he (rightly so) considers himself a journalist not a “pretty face” for TV.  We had to explain that it was mainly about keeping those eyes in focus on his slightly moving head that was the distraction.  The still below is a crop from the 4K image.

cu wide Just look and you can see the focus fall off from the bridge of the nose to his sideburns.  In fact it appears that I have the front of his eyeball in focus but the edge, being a sphere, was just that much further away that it was not in focus. and as I was shooting 4K or 4 times the resolution of HD there was nowhere to hide.  you had it or you didn’t.

eye ecuWe got through the day, but boy, it was a wake-up call for how hard even the simplest job can bring unexpected challenges.  No dolly shot no intentional talent movement, yet focus was the biggest challenge of the day.  We weren’t even shooting wide open.

Moral to the story: be careful what you ask for.  There can be a thing as too shallow a depth of field.  Always ask for an assistant. And 4K, and beyond can be very, very unforgiving if you get it wrong. And maybe Victorian head clamps may come back in to style.